Skip to content

fix for empty string input to openai/text-embedding-3-large #2634

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
ayush1298 wants to merge 4 commits into from
Closed

fix for empty string input to openai/text-embedding-3-large #2634

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
7dabf91
fix for empty string input to openai/text-embedding-3-large
ayush1298 May 3, 2025
9866dac
added correct fix
ayush1298 May 4, 2025
2f9ec43
added suggestions
ayush1298 May 5, 2025
aaddce3
Remove unnecessary check
ayush1298 May 7, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 19 additions & 5 deletions mteb/models/openai_models.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,9 +18,10 @@ class OpenAIWrapper(Wrapper):
def __init__(
self,
model_name: str,
*,
max_tokens: int,
embed_dim: int,
tokenizer_name: str = "cl100k_base", # since all models use this tokenizer now
embed_dim: int | None = None,
**kwargs,
) -> None:
"""Wrapper for OpenAIs embedding API.
Expand Down Expand Up @@ -54,17 +55,23 @@ def truncate_text_tokens(self, text):
return self._encoding.decode(truncated_sentence)

def encode(self, sentences: list[str], **kwargs: Any) -> np.ndarray:
requires_package(self, "openai", "Openai text embedding")

print("Starting OpenAIWrapper encode method")
from openai import NotGiven

empty_mask = [not s.strip() for s in sentences]
if self._model_name == "text-embedding-ada-002" and self._embed_dim is not None:
logger.warning(
"Reducing embedding size available only for text-embedding-3-* models"
)

if any(empty_mask):
logger.warning(
"Empty strings detected - encoding non-empty only, returing zero vectors for empty strings."
)

non_empty_sentences = [s for s, e in zip(sentences, empty_mask) if not e]
trimmed_sentences = []
for sentence in sentences:
for sentence in non_empty_sentences:
encoded_sentence = self._encoding.encode(sentence)
if len(encoded_sentence) > self._max_tokens:
truncated_sentence = self.truncate_text_tokens(sentence)
Expand Down Expand Up @@ -118,7 +125,11 @@ def encode(self, sentences: list[str], **kwargs: Any) -> np.ndarray:
)
all_embeddings.extend(self._to_numpy(response))

return np.array(all_embeddings)
all_embeddings = np.array(all_embeddings)
final_embeddings = np.zeros((len(sentences), self._embed_dim), dtype=np.float32)
Copy link
Copy Markdown
Contributor

@KennethEnevoldsen KennethEnevoldsen May 4, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

this already fill out the empty texts.

You can then simply fill in the text_embeddings using:

mask = [i for i, t in enumerate(text) if t.strip()]
final_emb[mask, :] = text_emb

Here is a small sample to show the idea:

import numpy as np

matrix1 = np.zeros((5, 4))
matrix2 = np.random.rand(3, 4)
matrix1[[0, 2, 4],:] = matrix2

Copy link
Copy Markdown
Collaborator Author

@ayush1298 ayush1298 May 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Didn't get these one. Current implementation was 1st creating whole vector of zeros, and then only setting text embeddings, at positions where non-empty text is present. I think you are suggesting same only.

non_empty_idxs = [i for i, e in enumerate(empty_mask) if not e]
final_embeddings[non_empty_idxs] = all_embeddings
return final_embeddings

def _to_numpy(self, embedding_response) -> np.ndarray:
return np.array([e.embedding for e in embedding_response.data])
Expand All @@ -134,6 +145,7 @@ def _to_numpy(self, embedding_response) -> np.ndarray:
model_name="text-embedding-3-small",
tokenizer_name="cl100k_base",
max_tokens=8191,
embed_dim=1536,
),
max_tokens=8191,
embed_dim=1536,
Expand All @@ -159,6 +171,7 @@ def _to_numpy(self, embedding_response) -> np.ndarray:
model_name="text-embedding-3-large",
tokenizer_name="cl100k_base",
max_tokens=8191,
embed_dim=3072,
),
max_tokens=8191,
embed_dim=3072,
Expand All @@ -184,6 +197,7 @@ def _to_numpy(self, embedding_response) -> np.ndarray:
model_name="text-embedding-ada-002",
tokenizer_name="cl100k_base",
max_tokens=8191,
embed_dim=1536,
),
reference="https://openai.com/index/new-and-improved-embedding-model/",
max_tokens=8191,
Expand Down